# Agentic Robotics: What It Takes to Build a Tattoo Robot

November 2, 2026 · https://tatbot.ai/updates/2026-11-02/

On October 28, I brought tatbot, the open-source tattoo robot, to [Austin Robotics at HICAM](https://www.meetup.com/austin-robotics/events/312776081/). My talk, **Agentic Robotics: The Evolving Landscape of Agents and Policies**, explored how software agents, 3D tools and robot policies fit into the work of building physical machines.

[Browse the 36-slide public edition](/assets/updates/2026-11-02-agentic-robotics/index.html) or [download the slides as a PDF](/assets/updates/2026-11-02-agentic-robotics/austin-robotics-public.pdf).

## Software agents and robot policies

An agent needs more than a language model. Tools, context and a loop for acting on results give it a way to work on a task. In software, that might mean editing code, running it and reading the output.

Robotic control introduces a different job: turning observations into actions on hardware. The talk follows that distinction through visuomotor policies, vision-language-action models and systems that pair a planner with a robot policy. The System 1 and System 2 analogy helps describe these roles, although the cited projects implement them differently.

## 3D tools and a tattoo robot's workspace

The slides also explore agents for 3D work, including MeshFormer and Blender tools, alongside Tatbot's palette and hardware. A tattoo robot brings software decisions into a physical workspace: cameras observe surfaces, calibration relates those observations to the arm, and a tool must follow the intended path.

I brought the arm for a live hardware demonstration so the discussion had a physical reference point. The photographs above show the talk and the conversations around the robot afterward.

## The robotics research loop

The last part of the presentation looks beyond a single action: propose an experiment, run it, inspect the results and choose the next change. I call this outer research loop **System 3** in the talk.

Research examples include ASPIRE and ENPIRE. Their results belong to their authors' tasks and evaluation conditions. The Tatbot architecture at the end of the deck is a research direction; the talk does not establish autonomous tattooing capability.

## Explore the talk

The public slides include links to the original research and demonstrations. Video previews open their original sources. Thank you to Austin Robotics and HICAM for hosting; see [Tatbot's event photographs](https://x.com/tatbots/status/2115836274686808232) and explore the [open-source tattoo robot project](https://github.com/hu-po/tatbot).